I guess PPUs really never had a chance.

S

skilzygw

Newcomer
This comment comes from the fact that Intel announced Q4-06 release dates for their Quad Core processors. Say what you will about Intel's version being a kludge with just 2 conroe's slapped together. It should still be more than enough for physics processing. I really don't see people spending $300 for a ppu when these things are just around the corner and with Conroe's aggressive pricing scheme probably at a very affordable price.

I know this topic has been started before, but with the recent Intel announcement of an accelerated roadmap i think it is relevant.

Thanks.
 
Last edited by a moderator:
No. No no. No no no no no no no no no. It's not parallel enough to be fast enough in games. This is why you need to do GPGPU programming or have a dedicated vector processor like a PPU.
 
four x86/x64 fully out-of-order superscalar cores each with powerful SIMD, that begins to look very good and should kick Cell's ass for gaming. (yay, when cores are used but problem is the same on consoles)

And, contrary to PPU (which slows down games) you don't necessarily have to deal with API overhead, and don't have to sent data forth and back on a 133MB/s bus shared by other peripherals.

don't have great hopes about cheapness though, rather expect an eXXtreme price for that dual conroe. things will be interesting too against AMD 4x4, while K8 is less powerful than conroe it has NUMA architecture whereas Intel will have four cores fighting over the same FSB. with theoric numbers we're talking 25.6GB/s bandwith for AMD 4x4 and 10.6GB/s for Intel with a FSB 1333, or 8.5GB/s with FSB 1066 (more likely, if you look at xeon there's still FSB 800 and 667, guess running it high is harder with two physical CPU (if CPU = die))
 
I don't think PPU's in the form AGEIA have brought them to the market for PC have a chance for other reasons. Primarly it is an extremely niche market that cares about speed and when they see adding a PPU decreases their FPS they are put off and the fact that to make proper use of the PPU in a game would essentially require two different games... adding an expensive PPU for fancy graphical effects like cloth animation isn't what physics in games should be about. You only have to take a look at Monkey Ball and Marble Madness to see how physics can be the basis of your gaming idea.
 
Blazkowicz_ said:
four x86/x64 fully out-of-order superscalar cores each with powerful SIMD, that begins to look very good and should kick Cell's ass for gaming.
Click to expand...
It'll also cost as much as an entire PS3 and require a heatsink from hell. Yay. :rolleyes: It would also need games that actually are programmed for multithreaded processors, and with a precious few exceptions, those aren't exactly all that close on the horizon.

(yay, when cores are used but problem is the same on consoles)
Click to expand...
Not really. All 360s and PS2s are multicored/threaded. Only a minority of the market has hyperthreaded and/or dual-core CPUs, and just a tiny fraction of the PC market will have quad cores... Which platform do you target as a PC developer, I'm just asking? The one with 4 cores and virtually no systems installed in the marketplace, or the one with a single CPU?

and don't have to sent data forth and back on a 133MB/s bus shared by other peripherals.
Click to expand...
That's why I'm waiting for a PCIe PPU board. :p
 
there are clossplatform engines (UE3, next Carmack) where PC version and Xbox 360 version should be very similar IMO..
and a PPU costs as much as an entire HDD-less X360 :)
 
interesting when you say "four x86/x64 fully out-of-order superscalar cores each with powerful SIMD, that begins to look very good and should kick Cell's ass for gaming. (yay, when cores are used but problem is the same on consoles)"

is it not true that clock for clock that even sse4's limited 128bit capabilitys are still not as good as the Altivec when it comes to the best out of vectorising code?.

sure the x86 has the massive clock advantage, but dont you think someone would at least sit down and make a comparable vector unit, perhaps the PPU can fill that gap and in a very good way too?.

as for the Altivec enabled cell , i think its far to early to rubish it before they have even started to show what vector code can do.

theres a Cell tool chain ported to Gentoo if you have a notition to check it out http://www.ppczone.org/forums/viewtopic.php?t=682


a simulator too if you look hard enough.
 
popper said:
is it not true that clock for clock that even sse4's limited 128bit capabilitys are still not as good as the Altivec when it comes to the best out of vectorising code?.
Click to expand...
This used to be true for the anemic half-width SSE units present in Pentium4 and Athlon64, but the SSE units in Conroe are a rather massive upgrade from those - I would be quite surprised if any current Altivec implementation can match the clock-for-clock performance of THREE full-width 128-bit SSE units running in parallel (which, after all, is what Conroe has per core).
 
SSE3 already brought the vertical register op's capability and now Conroe (and K8L) single-cycle 128 bit superscalar processing, I think the AltiVec advantages are well suppressed, where may be the lack of more FP register space is something that SSE is still lacking (from a compiler point of view).
 
You must log in or register to reply here.
Back
Top